Digital all-sky polarization imaging of partly cloudy … sky polarization...cloud in a partly...

Digital all-sky polarization imagingof partly cloudy skies

Nathan J. Pust and Joseph A. Shaw*Department of Electrical and Computer Engineering, 610 Cobleigh Hall,

Montana State University, Bozeman, Montana 59717, USA

*Corresponding author: [email protected]

Received 4 June 2008; accepted 6 August 2008;posted 13 October 2008 (Doc. ID 96960); published 19 November 2008

Clouds reduce the degree of linear polarization (DOLP) of skylight relative to that of a clear sky. Even thinsubvisual clouds in the “twilight zone” between clouds and aerosols produce a drop in skylight DOLP longbefore clouds become visible in the sky. In contrast, the angle of polarization (AOP) of light scattered by acloud in a partly cloudy sky remains the same as in the clear sky for most cases. In unique instances,though, select clouds display AOP signatures that are oriented 90° from the clear-sky AOP. For theseclouds, scattered light oriented parallel to the scattering plane dominates the perpendicularly polarizedRayleigh-scattered light between the instrument and the cloud. For liquid clouds, this effect may assistcloud particle size identification because it occurs only over a relatively limited range of particle radii thatwill scatter parallel polarized light. Images are shown from a digital all-sky-polarization imager to illus-trate these effects. Imagesare also shown that provide validation of previously published theories forweak(∼2%) polarization parallel to the scattering plane for a 22° halo. © 2008 Optical Society of America

OCIS codes: 120.5410, 260.5430, 290.1310, 280.0280.

1. Introduction

For clear skies, molecular (Rayleigh) scattering inthe atmosphere polarizes visible light in directionsperpendicular to the scattering plane—the planethat contains the Sun, the scattering object, andthe observer. Previous and present research observesthat cloud presence reduces the measured degree oflinear polarization (DOLP) from that of the clear sky[1–3]. This DOLP reduction is caused primarily bymultiple scattering within the cloud, which depo-larizes the light. In spite of the reduced DOLP, inpartly cloudy skies the angle of polarization (AOP)of light scattered from clouds often remains consis-tent with that of the clear sky [1,2,4]. This measuredcloud AOP is not produced by the cloud itself, but in-stead represents the AOP of the polarized Rayleigh-scattered light in the path between the cloud and theobserver—while light scattered from the cloud itselfis effectively unpolarized.

Previous investigators have focused upon the ob-served consistencies between measured cloud AOPand clear-sky AOP and did not focus on observed de-viations of cloud AOP from the clear-sky pattern [2].However, we have observed many instances wherecloud AOP exhibits departures from the clear-skyAOP. In this paper we present observations madeunder clear and partly cloudy conditions and identifyseveral interesting cases where the images deviatefrom what might be expected from simple considera-tions of clear and cloudy conditions. In the process,we show what we believe is the first polarimetricevidence of a region of enhanced aerosol separatingclouds from the clear sky, recently dubbed the“twilight zone” [5]. We also use a polarization imageto provide support for previously published theoriesdescribing the polarization state of light in a 22° halo.

2. Polarization Images of Clear and Cloudy Skies

We previously described a dual-field, imaging polari-meter designed for studying both sky polarizationand ground-based object polarization signatures

0003-6935/08/34H190-09$15.00/0© 2008 Optical Society of America

H190 APPLIED OPTICS / Vol. 47, No. 34 / 1 December 2008

at visible and near-infrared wavelengths [1]. Thispolarimeter can be switched between two fields ofview—a wide-angle fisheye field of view for imagingthe full sky and a narrow field of view for imagingsmaller objects. This instrument operates in five10nm bands defined by interference filters centeredat 452, 491, 532, 632, and 701nm. Two liquid-crystalvariable retarders (LCVRs) are used to electronicallyvary the polarization sensitivity of the imager so thata full Stokes image is measured in less than a fewtenths of a second at each wavelength. The resultingset of four intensity images from a CCD camera areinverted by a calibration matrix to form 4-elementStokes images [1,6]. This speed allows reliable mea-surements in partly cloudy skies without polariza-tion artifacts that would arise when clouds movemore than a small fraction of a pixel betweenframes.The dual-field polarization imager used here builds

upon and extends a varied legacy of sky-polarizationimagers.NorthandDuggin [7] useda four-lens stereo-scopic camera pointed down at a dome mirror torecord all-sky polarization images from scanned film.Shortly thereafter, Lee [8] published an insightful dis-cussion of sky-polarization images obtained fromscanned film exposed through a manually rotatedlinear polarizer. Pomozi et al. [2] and Horvath et al.[3] have performed a wide variety of studies withan all-sky polarization imager that uses scanned filmimages recorded through linear polarizers. Voss andLiu [9] developed adigital all-sky-polarization imagerthat used rotating filter wheels to vary both polariza-tion state andwavelength. Liu andVoss [10] used thissystem to explore the variation of clear-sky polariza-tion with atmospheric optical depth and surface re-flectivity. Recently, Cronin et al. [11] used a digitalall-sky-polarization imager with a manually rotatedlinear polarizer to study sky-polarization patternsduring twilight. Our liquid-crystal-based imager pro-vides the advantages of previous digital imagers, and

also provides a speed advantage over systemsthat rely on time-sequential polarization images.The subsecond time required to gather a full setof four polarization images allows this system tomeasure sky polarization with variable cloudswithout polarization artifacts that arise when cloudsmove between images [1,6,11]. In the present studythe instrument is used to study clear and partlycloudy sky polarization.

Figure 1 shows the 450nm degree of linear polar-ization (DOLP) and angle of polarization (AOP), re-spectively, for a clear sky seen through the fisheyelens of our LCVR-based imaging polarimeter at17:55 (Mountain Standard Time, MST ¼ UTC−6h) on 26 June 2006 from the top of the engineeringbuilding at Montana State University in Bozeman,Montana (45:67 °N, 111:05 °W). In the DOLP image,the band of maximum polarization caused byRayleigh scattering is clearly visible in an arcpositioned 90° from the Sun (with a maximumDOLP ¼ 66%). Similar data are seen in the otherfour wavelength bands (not shown), with the longerwavelengths exhibiting higher maximum DOLPvalues. The oval at the top of the image is the Sunocculter, which prevents overexposure by direct sun-light. Also, penthouses on the top of the building canbe seen to the top (west) and the top left (southwest).The black area below the occulter in the AOP imageis the Babinet neutral point. This area has been de-noted black because the AOP is undefined where theDOLP is equal to zero. The other black area to theright of the occulter is an area of overexposed reflec-tion from a building antenna. It should be noted thatAOP values in regions away from the image centerare difficult to interpret because the fisheye lensprojects a three-dimensional polarized sky dome ontoa two-dimensional Stokes image. For reference,though, a 90° AOP measurement at the image centerrepresents a vector oriented from left to right.

Fig. 1. Clear-sky polarization at 450nm (26 June 2006, 5:55 pm MDT ¼ UTC − 6h). Black areas denote regions that are overexposed,underexposed, uncalibrated, or—in the case of the AOP—undefined due to lack of measurable polarized light. For AOP to be considereddefined, a threshold of DOLP > 1% was chosen.

1 December 2008 / Vol. 47, No. 34 / APPLIED OPTICS H191

A. Aerosol and Subvisual Cloud Influence on Polarizationin Apparently Clear Sky

Figure 2 shows an image of a single cloud in an other-wise clear sky. The image was taken at the same timeof day (early evening) as Fig. 1—but 3 days earlier.The DOLP is considerably lower for the cloud thanfor the clear sky, owing to the largely unpolarizedlight arising from multiple scattering in the cloud.The clear-sky DOLP is also reduced in this imagecompared to Fig. 1. While the exact cause of this re-duction remains to be conclusively determined, theeffect is consistent in images containing clouds. Itis our feeling that unseen aerosols and possibly thinclouds in what has recently been called the “twilightzone” between a cloud and the clear sky [5,12,13] arereducing the DOLP in what appears to be clear sky.We believe that this effect on the sky polarization isdirectly related to the recently described observa-tions of enhanced optical depth near clouds [5]. Inpartially cloudy skies, we see DOLP reductions inclear-sky areas between clouds that appear to becaused by subvisual aerosols and/or clouds. (Eventhough clouds appear to have hard edges, they arein fact surrounded by thin clouds.) Furthermore,these DOLP reductions show up in the clear sky longbefore we can physically see clouds in the sky.Figure 3 shows this effect at 700nm, but a very

similar effect is seen at all of our observed wave-lengths. The data from 11 September and 26 Junedemonstrate the typical progression of the maximumDOLP through a day with a clear sky. However, inthe data from 23 June the maximum DOLP beginsdropping away from the clear-sky curve around thesolar elevation maximum (65:5° at 13:30 MST).Not until 3h later (solar elevation angle ¼ 41° at16:30) are visible clouds seen in the imagery. Webelieve this drop in DOLP is caused by subvisualaerosols/clouds during cloud formation. Also, thelong length of time (3h) between the first DOLP dropand the appearance of visual clouds—indicated by

the gray area in Fig. 3—gives additional supportto the recent claim that such subvisual clouds stretchtens of kilometers away from the visual clouds [5].This effect is also seen after visible clouds leavethe scene, indicated in the area following the gray-shaded period in Fig. 3. Therefore, we believe thepolarimeter is seeing thin aerosol/cloud droplets thatare not visible to the naked eye before and afterclouds become visible to the eye.

In Fig. 2, it is notable that the cloud AOP does notdeviate from the clear-sky AOP. This is typical whenthe clouds move through the band of maximumDOLP that is 90° from the Sun and when otherclouds do not block the Sun. Nevertheless, many in-stances of cloud AOPs that depart from the clear skyhave also been measured.

B. Observed Cloud AOPs that Depart from the Clear-SkyPolarization Pattern

Figure 4 shows an example of clouds that deviatefrom the AOP of the clear sky. Again, this image istaken at a similar afternoon time to Figs. 1 and 2—but a month earlier. As expected, the AOP inclear-sky areas between the clouds is similar toFigs. 1 and 2, but the AOP of several clouds in theimage deviate from the clear-sky AOP. (Similar dataare seen in 490, 530, and 630nm images, but are notseen in the 450nm image.) We assume that this is aliquid cloud based on its appearance. A regionalradiosonde profile from Great Falls, Montana, sug-gested a cloud-base temperature between 0 and−5 °C, which is certainly reasonable for a liquidcloud. Also, for the 55–90° scattering angles of thiscloud, single scatter models have consistently shownthat cirrus ice crystals exhibit polarization anglesthat are in the same direction as for a clear sky—thatis they are polarized perpendicular to the scatteringplane [14–17]. If the bottom of the cloud was ice, itsAOP would not be expected to follow the pattern dis-cussed here.

Fig. 2. Cloud polarization in a partially cloudy sky at 450nm (23 June 2006, 5:55 pmMDT ¼ UTC − 6h). Black areas denote regions thatare overexposed, underexposed, uncalibrated, or—in the case of the AOP—undefined due to lack of measurable polarized light. For AOP tobe considered defined, a threshold of DOLP > 1% was chosen.


Several observations of Fig. 4 should be noted.First, the AOP throughout the image is either (1)identical to the clear-sky AOP, (2) oriented 90° fromthe clear-sky AOP, or (3) undefined. Up to the time ofpublication, this remains true for all data obtainedby this imager. The AOPs have always remainedwithin these three cases. Second, the clouds insideor near the band of maximum DOLP—at the bottomleft in this image—exhibit little change from theclear-sky AOP. This is consistent in most images.It is rare for a cloud to show a change in the AOPwhen it is located in the high-DOLP band. Finally,the image shows many clouds in the vicinity of theSun (behind the occulter). The Sun is most certainlypartially blocked by these clouds. In all data taken bythe imager, blocking of the Sun by clouds highly cor-relates to instances of cloud AOP changing from the

clear-sky AOP. Therefore, more clouds show changesin AOP relative to the clear sky when either (1) theSun is obscured by clouds, or (2) the clouds arelocated in sky regions away from the band of highclear-sky DOLP. Other researchers have seen similareffects [2]. Reasons for each of these observations willbe discussed in the next section.

3. Scattering Path Contributions to MeasuredPolarization

To better understand cloud AOP deviations frommolecular-scattering AOP in a partly cloudy sky, it isimportant to consider individual contributions fromthe scattering path to the polarimetric imager.Figure 5 shows the independent scattering mechan-isms that contribute to the polarimetric image,assuming an unpolarized source. First, Rayleigh

Fig. 3. Maximum degree of polarization for 23 June, 26 June, and 11 September 2006 at 700nm. The sky was completely cloud free on 26June and 11 September, while on 23 June the sky was perceived to be cloud free except in the gray-shaded time when a few small, localizedclouds were visible (see Fig. 2). The flat spot in the graph of 11 September data occurs when the Sun never attained that height. The extrax-axis hash shows the maximum solar elevation for 23 June.

Fig. 4. Polarization for a cloud that changes the AOP at 700nm (24 May 2006, 5:21 pm MDT ¼ UTC − 6h). Black areas denote regionsthat are overexposed, underexposed, uncalibrated, or—in the case of the AOP—undefined due to lack of measurably polarized light. ForAOP to be considered defined, a threshold of DOLP > 1% was chosen.


scattering below the cloud (denotedA in Fig. 5) intro-duces light into the path. This light contains a com-ponent that is polarized perpendicular to thescattering plane—the plane including the Sun, thescattering object, and the observer—and a compo-nent that is unpolarized. (For Fig. 5, the scatteringplane is the plane of the paper.)Second, the polarized component of the light from

the cloud consists of light that is both singly andmul-tiply scattered (denoted B in Fig. 5). Single-scatteredlight from liquid clouds will be polarized eitherperpendicular or parallel to the scattering plane—depending upon cloud parameters such as drop sizeand cloud optical depth. Previous research hasshown that most polarization from the cloud arisesfrom single scattering, while multiply scattered lightprimarily reduces the DOLP [18]. Models haveshown that, in the vast majority of cases, the AOPof the multiply scattered light is the same as theAOP of the singly scattered light [19,20]. Therefore,the polarized component of the light scattered by thecloud nominally will be polarized either parallel orperpendicular to the scattering plane.Third, an unpolarized component of multiple scat-

tering emanates from the cloud (denotedC in Fig. 5).Although this component is not strictly physicallyseparate from the polarized component of the cloudradiance, conceptually it is helpful to consider itindependently from the polarized component.Finally, perpendicularly polarized Rayleigh

scattering above the cloud (denoted D in Fig. 5)may contribute to the measurement. For thickclouds, it is not necessary to consider the molecular

scattering above the cloud because multiple scatter-ing in the cloud destroys it. Still, for thin clouds itmay contribute significantly to the polarimetricimage.

The DOLP and AOP of the measured light dependupon the relative magnitudes of each of thesecomponents. The following paragraphs describe fivecommon cases:

1. The unpolarized light frommultiple scatteringinside the cloud (C) severely dominates the bright-ness of the other components. Any polarized compo-nent from the other contributors (A, B, and possiblyD) is so small that it remains below the detectablelimit of the instrument. In this case, the instrumentwill measure unpolarized light (DOLP ¼ 0) with anundefined AOP.

2. If the polarized component of the cloud scatter-ing (B) is perpendicular to the scattering plane, thenthe polarimeter will only see polarization that isoriented perpendicular to the scattering plane (fromcomponents A, B, and possibly D). For this case, themeasured DOLP will be reduced from the clear-skyvalue by the unpolarized component from the cloud(C), while the AOP of the polarimetric image will bein the same direction as the clear-sky AOP.

3. If the polarized component of the cloud scatter-ing (B) is polarized parallel to the scattering plane,but is dimmer than the perpendicularly polarizedRayleigh scattered components (A and D), thenthe polarized component of the cloud will reducethe polarized component of the Rayleigh scattering.The polarized component of the cloud effectivelydepolarizes the stronger perpendicularly polarizedRayleigh scattering. In this case, the polarimetermeasures a clear-sky DOLP component that is re-duced by both the polarized component of the cloudand the unpolarized component from the cloud, butthe AOP remains oriented in the same direction asthe light scattered from the surrounding clear sky.

4. If the polarized component of the cloud scatter-ing (B) is parallel to the scattering plane, and is equalin brightness to the Rayleigh scattered componentsthat are polarized perpendicular to the scatteringplane, then the polarized Rayleigh components (Aand D) and the polarized cloud component (B) willcancel each other. The imager will measure comple-tely unpolarized light.

5. Finally, if the polarized component of the cloud(B) is parallel to the scattering plane, but is brighterthan the polarized Rayleigh-scattered componentsabove and below the cloud, then the parallel-polarized cloud polarization will dominate the polar-ization of the Rayleigh scattering above and belowthe cloud (A and D). In this case, the AOP will bemeasured parallel to the scattering plane—or 90°from the clear-sky case, while the instrument willmeasure a cloud DOLP that is reduced by both theperpendicularly polarized Rayleigh scattering aboveand below the cloud and the unpolarized componentof light scattered by the cloud.

Fig. 5. Scattering contributors to the polarimetric groundmeasurement. The plane of the paper is the scattering plane.The polarized portion of the Rayleigh scattered components (Aand D) is always polarized perpendicular to the scattering plane,while the polarized component of theMie scattering from the cloudcan potentially be polarized either parallel or perpendicular tothe scattering plane. θ is the scattering angle. (θ ¼ 0 is forwardscattering.)


It should be apparent that for many observationswhere clouds exhibit the same polarization as theclear sky, the observed AOP does not arise fromthe cloud itself but instead is produced by theRayleigh scattered light in the atmospheric columnbetween the observer and the cloud.For all the above cases, the AOP is theoretically

either the same as the clear sky, 90° from the clearsky, or undefined. The observed AOP depends uponthe relative amplitudes of each scattering compo-nent. For cases where the Rayleigh scattered compo-nents dominate, the AOP is consistent with that ofthe clear sky. For cases where the polarized cloudcomponent dominates, the AOP can at times be90° from the clear sky (depending upon the polariza-tion direction of the cloud droplet scattering).Case 5 explains why it is more common to observe

AOP changes from the clear sky when either the Sunis blocked by clouds, or the clouds are positionedaway from the band of increased clear-sky DOLP.In these cases, either the total Rayleigh scatteringbelow the cloud is reduced—as when the Sun isblocked—or the polarized Rayleigh scattering islow—as when the cloud is away from the maximumDOLP band. Both cases reduce the polarizedRayleigh scattered light in the column below thecloud, allowing the parallel-polarized cloud light tobecome brighter than the perpendicularly polarizedcomponent of the Rayleigh scattered light. In thesecases, a measurable change in AOP of the cloud fromthat of the clear sky occurs. In contrast, it is rare for acloud to show a change in AOP from the clear skywhen either the column below the cloud is fully litby the Sun or the cloud is near the maximum

clear-sky polarization band. In these cases the polar-ized Rayleigh scattered component has a largermagnitude—and is thus more difficult to overcomeby the polarized cloud component (Case 3).

Case 5 describes the unique conditions that mustexist for measured AOP of a cloud to be differentthan the clear sky. In this case, the polarized com-ponent of cloud scattering is larger than the polar-ized Rayleigh scattering. This case predicts that theAOP will be oriented 90° from the surroundingclear-sky AOP. As can be seen in Fig. 6, the AOPimagery agrees perfectly with this prediction. Thereis clearly a 90° shift between the cloud and the sur-rounding clear sky.

It should be noted that this treatment ignores thecase of secondary illumination of both the cloud andthe column beneath the cloud by other surroundingclouds. If the Sun was adequately blocked and an-other cloud illuminated the column, the AOP couldpotentially change (this would be a special case ofmultiple scattering). If this were the case, the AOPwould shift from one of the three expected values(0° from the clear-sky AOP, 90° from the clear-skyAOP, or undefined). Since all data observed thusfar have shown AOPs that fall within the threeexpected values, we are confident that the issue ofsecondary illumination is sufficiently small to beneglected in this treatment.

4. Cloud Droplet Size Estimation

For liquid clouds that change the AOP, it is clear thatthe scattering processes in the cloud scatter lightthat is predominantly polarized parallel to the scat-tering plane. Our data suggest that—although it is

Fig. 6. Areas of spherical particle sizes that scatter a polarization that is parallel to the scattering plane (shown in the gray areas). Themode radius of the cloud droplet size distribution (the most frequent radius size of the distribution) is labeled on the x axis. (Distributionparameter μ ¼ 20.)


difficult to obtain a magnitude of the scattering fromthe cloud—the polarization angle of the scattering it-self may provide information about the droplet sizedistribution composing the cloud. Although research-ers have used polarization from satellite and air-borne measurements to retrieve liquid cloudparameters [21–23], to our knowledge this has notbeen done with ground-based sky-polarization ima-ger data. To demonstrate the feasibility of this tech-nique, we assumed that the polarization primarilyarises from single scattering within the cloud andran several single-scatter Mie models to find dropletsizes that scattered light polarized parallel to thescattering plane.For each of the bands of the imager, Mie scattering

results were obtained over 5 wavelength samplesacross the band and then averaged to account forwavelength dispersion in the scattering. The indexof refraction of the water was taken to be n ¼ 1:33.The cloud droplet distribution function was selectedto be a gamma distribution following Kokhanovsky[18]. The parameter μ was selected to be 20, 10, 5,and 1. (This parameter is inversely related to thewidth of the droplet distribution. μ ¼ 20 representsa narrow distribution, while μ ¼ 1 is a wide distribu-tion.) The number of particles used was 50. Theywere sequentially selected from a random gammadistribution of 1000 that had been sorted. (This en-sured that the distribution was representative, butstill had some random attribute.)For each wavelength, a mapping of droplet distri-

butions that scattered a parallel polarization wascalculated for scattering angles from 20 to 160°.Since the AOP deviation from the clear sky in Fig. 4was observed not only in the 700nm band, but also at630, 530, and 490nm (not shown), cloud droplet radii

that would scatter parallel polarization for all ofthese 4 bands were mapped. The areas where thedroplet distributions would scatter parallel polariza-tion are shown as gray in Fig. 6 (for a distribution ofμ ¼ 20). The mode of the distributions (mostfrequent drop size) is plotted on the x axis.

Variation in the width of the distribution functiondid not appreciably affect the data except that largedistributions further limited the set of particle sizesthat could scatter parallel polarization. In otherwords, a narrow cloud droplet size distribution canscatter parallel polarization for a larger set of meanparticle sizes than a wide distribution. To include themost possible size distributions, the μ ¼ 20 plot isshown in Fig. 6. (To show that the dependence onμ is relatively insignificant, Fig. 7 is also providedfor μ ¼ 5.)

Figure 6 shows that for small scattering angles adeparture of measured cloud AOP from the clear-skyAOP does not give any useful information aboutmean droplet size. Many different droplet distribu-tions can give the parallel AOP signature. Still, forlarge scattering angles (80–130°), an AOP departurefrom the clear-sky value indicates that the meandrop size must be below 7 μm. In the example datashown in Fig. 4, only the green colored clouds on thefar right of the image would have scattering anglesthis small. The scattering angles for these cloudswould be roughly 80–90°. Therefore, the mode clouddroplet size for this particular cloud must be be-tween 0.5 and 7 μm. This method shows potentialto provide auxiliary cloud droplet information. Still,further investigation with other instruments isneeded to validate this method for a larger clouddata set.

Fig. 7. Same as Fig. 6 but with a wider cloud droplet distribution function (μ ¼ 5).


5. Halo Polarization

On 24 April 2006 we obtained polarimetric imagesof a 22° halo within a sequence of cloud images.Figures 8 and 9 show images taken of the halo at450nm. (The Sun occulter covers the Sun.) Althoughthe halo was not complete in any of the images taken,these images show similar results to previous re-searchers [24–26]. These data provide further confir-mation of previous theoretical predictions andmeasurement, which show that the 22° halo is polar-ized parallel to the scattering plane [24,25]. Like theclouds shown in Fig. 4, the halo AOP is parallel to theclear sky behind it. This is shown in Fig. 9 and otherwavelengths show similar AOP. For many images,though, the DOLP became low and noisy due to var-iation in brightness from the cloud structure in theimage. Even so, none of the halo data exhibitedDOLPs greater than 2%. This was consistent withprevious investigators [24,25]. It should be notedthat the method used previously to constrain theliquid cloud droplet sizes (shown in Fig. 4) cannotbe used for the halo since in this case the cloud par-ticles are ice crystals.

6. Conclusion

We have shown examples of clear and cloudy sky po-larization images measured with a fisheye imagerthat is capable of recording full digital Stokes-vectorsky images rapidly enough to avoid polarization er-rors in rapidly changing sky conditions. The imagesshow that there is significantly reduced maximumDOLP in apparently clear regions of sky when evenone small cloud is present elsewhere in the sky, rela-tive to what is observed in a truly clear sky. Thiseffect is observed for multiple hours before a cloudbecomes visible and for multiple hours after a clouddisappears, lending support to recent reports of high-er optical depth from aerosols and subvisual cloudsin what has been called the “twilight region” betweenclouds where the sky is not truly clear.

While many of our measurements show that theAOP is unchanged beneath a cloud, we have alsoshown images where the cloud significantly changesthe AOP from that of the immediately adjacent clearsky. Whether a cloud changes or retains the clear-skyAOP depends on the polarization state of light scat-tered from the cloud (i.e., whether the cloud is com-posed of liquid or ice) and on the relative brightnessof the light scattered by the cloud, the light trans-mitted through the cloud, and the light scatteredbelow the cloud. For liquid clouds this effect can beused to assist in cloud droplet size retrieval.

Finally, images of a 22° halo provide validation ofpreviously published theories for halo polarization.Our measurements show that the halo is linearlypolarized parallel to the scattering plane, with aDOLP of 2% or less.

This material is based on research sponsored bythe Air Force Research Laboratory, under agreementnumber FA9550-07-1-0011. The U.S. Government isauthorized to reproduce and distribute reprints forGovernmental purposes notwithstanding any copy-right notation thereon. The views and conclusionscontained herein are those of the authors and should

Fig. 8. Intensity image of a 22° halo at 11:24:13 MDT ¼ UTC − 6on 12 April 2006 (450nm).

Fig. 9. DOLPand AOP images of a 22° halo at 11:24:13MDT ¼ UTC − 6 on 12 April 2006 (450nm). Areas of lowDOLP (1%) aremasked inblack. Black areas denote regions that are overexposed, underexposed, uncalibrated, or—in the case of the AOP—undefined owing to thelack of measurably polarized light. For AOP to be considered defined, a threshold of DOLP > 1% was chosen.


not be interpreted as necessarily representing the of-ficial policies or endorsements, either expressed orimplied, of the Air Force Office of Scientific Researchor the U.S. Government.

References1. N. J. Pust and J. A. Shaw, “Dual-field imaging polarimeter

using liquid crystal variable retarders,” Appl. Opt. 45,5470–5478 (2006).

2. I. Pomozi, G. Horvath, and R. Wehner, “How the clear-skyangle of polarization pattern continues underneath clouds:full-sky measurements and implications for animal orienta-tion,” J. Exp. Biol. 204, 2933–2942 (2001).

3. G. Horvath, A. Barta, J. Gal, B. Suhai, and O. Haiman,“Ground-based full-sky imaging polarimetry of rapidly chan-ging skies and its use for polarimetric cloud detection,” Appl.Opt. 41, 543–559 (2002).

4. M. L. Brines and J. L. Gould, “Skylight polarization patternsand animal orientation,” J. Exp. Biol. 96, 69–91 (1982).

5. I. Koren, L. A. Remer, Y. J. Kaufman, Y. Rudich, andJ. V. Martins, “On the twilight zone between clouds and aero-sols,” Geophys. Res. Lett. 34, L08805 (2007).

6. J. S. Tyo, D. L. Goldstein, D. B. Chenault, and J. A. Shaw,“Review of passive imaging polarimetry for remote sensingapplications,” Appl. Opt. 45, 5453–5469 (2006).

7. J. A. North and M. J. Duggin, “Stokes vector imaging of thepolarized sky-dome,” Appl. Opt. 36, 723–730 (1997).

8. R. L. Lee, Jr., “Digital imaging of clear-sky polarization,” Appl.Opt. 37, 1465–1476 (1998).

9. K. J. Voss and Y. Liu, “Polarized radiance distribution mea-surements of skylight. I. System description and characteriza-tion,” Appl. Opt. 36, 6083–6094 (1997).

10. Y. Liu and K. J. Voss, “Polarized radiance distributionmeasurement of skylight. II. Experiment and data,” Appl.Opt. 36, 8753–8764 (1997).

11. T.W. Cronin, E. J.Warrant, andB. Greiner, “Celestial polariza-tionpatternsduring twilight,”Appl.Opt.45, 5582–5589(2006).

12. A. Marshak, G.Wen, J. A. Coakley, Jr., L. A. Remer, N. G. Loeb,and R. F. Cahalan, “A simple model for the cloud adjacencyeffect and the apparent bluing of aerosols near clouds,” J. Geo-phys. Res. 113, D14S17(2008), doi:10.1029/2007JD009196.

13. I. Koren, L. Oreopoulos, G. Feingold, L. A. Remer, andO. Altaratz, “How small is a small cloud,” Atmos. Chem. Phys.Discuss. 8, 6379–6407 (2008).

14. P. Goloub, M. Herman, H. Chepfer, J. Riedl, G. Brogniez,P. Couvert, and G. Seze, “Cloud thermodynamical phaseclassification from the POLDER spaceborne instrument,”J. Geophys. Res. 105, 14747–14759 (2000).

15. Y. Takano and K. N. Liou, “Radiative transfer in cirrus clouds.Pt. III: Light scattering by irregular ice crystals,” J. Atmos.Sci. 52, 818–837 (1995).

16. A. Macke, J. Mueller, E. Raschke, “Single scattering propertiesof atmospheric ice crystals,” J. Atmos. Sci. 53, 2813–2825 (1996).

17. L. C. Labonnote, G. Brogniez, J. C. Buriez, and M. Doutriaux-Boucher, “Polarized light scattering by inhomogeneoushexagonal monocrystals: validation with ADEOS-POLDERmeasurements,” J. Geophys. Res. 106, 12139–12155 (2001).

18. A. Kokhanovsky, “Optical properties of terrestrial clouds,”Earth Sci. Rev. 64, 189–241 (2004).

19. G. W. Kattawar and G. N. Plass, “Degree and direction of po-larization of multiple scattered light. 1: homogeneous cloudlayers,” Appl. Opt. 11, 2851–2865 (1972).

20. K. N. Liou, An Introduction to Atmospheric Radiation(Academic, 1980).

21. F. M. Breon and P. Goloub, “Cloud droplet effective radiusfrom spaceborne polarization measurements,” Geophys. Res.Lett. 25, 1879–1882 (1998).

22. A. A. Kokhanovsky, “The determination of the effective radiusof drops in water clouds from polarization measurements,”Phys. Chem. Earth B 25, 471–474 (2000).

23. W.G.Egan, S. Israel,M. Sidran,E. E.Hindman,W.R. Johnson,and V. S. Whitehead, “Optical properties of continentalhaze and cumulus and orographic clouds based on space shut-tle polarimetric observations,” Appl. Opt. 32, 6841–6852(1993).

24. G. P. Konnen, “Polarization and intensity distributions ofrefraction halos,” J. Opt. Soc. Am. 73, 1629–1640 (1983).

25. G. P. Konnen and J. Tinbergen, “Polarimetry of a 22° halo,”Appl. Opt. 30, 3382–3400 (1991).

26. D. Lynch, “Polarization models of halo phenomena I. Theparhelic circle,” J. Opt. Soc. Am. 69, 1100–1102 (1979).


Digital all-sky polarization imaging of partly cloudy … sky polarization...cloud in a partly...

Documents

Transcript of Digital all-sky polarization imaging of partly cloudy … sky polarization...cloud in a partly...